1. Read/Write Images
2. Visualization of Images
3. Inhomogeneity/Bias Field Correction
4. Skull Stripping/Brain Extraction
5. Image Registration
6. Tissue-Class Segmentation
7. Image operations
8. Complex Modeling (yeah - it's R)

• oro.dicom - read/write DICOM data, the nifti object
  • Pure R implementation
• dcm2niir - uses dcm2nii from Chris Rorden
  • Can handle Par/Rec and many scanner types
• matlabr - could use dicomread matlab code and excecute through R

Medical Imaging Task View

• oro.nifti: read/write data, the nifti object
• fslr: process data (need FSL for most of the functionality)
• ANTsR: process data (full toolbox)
• extrantsr: makes ANTsR work with nifti objects

• dti - adaptive smoothing and diffusion tensor tools

• fmri - post-processing analysis: linear models and p-value smoothing
• AnalyzeFMRI - fMRI analysis (last updated in 2013)

• spm12r package calls out MATLAB using SPM

Multi-modal dataset from HAMMER, (NIfTI conversion from ANALYZE).

• Data from https://www.nitrc.org/frs/?group_id=187 (testing folder in White_Matter_Lesion_Segmentation_Testdata.zip)

• 4 MRI sequences: T1-weighted, T2-weighted, PD, FLAIR

files
            t1             t2             pd          flair 
   "T1.nii.gz"    "T2.nii.gz"    "PD.nii.gz" "FLAIR.nii.gz" 

fslr: readnii uses oro.nifti::readNIfTI:

library(fslr)
base_t1 = readnii(files["t1"])

• like an array
• ANTsR uses pointers (faster), but not as intuitive

library(fslr)
fslr::ortho2(base_t1)

over_50 = mask_img(base_t1, base_t1 > 40); ortho2(base_t1, over_50)

image(base_t1, z = 55, plot.type = "single")

over_50[over_50 <= 0] = NA; over_50 = cal_img(over_50)
overlay(base_t1, over_50, z = 55, plot.type = "single")

ANTsR/extrantsr

• bias_correct from extrantsr package calls ANTsR::n4BiasFieldCorrection (N. J. Tustison et al. 2010)
• EM-like, assumes bias is smooth over space, logs the data

library(extrantsr)
n4_t1 = bias_correct(file = base_t1, correction = "N4", retimg = TRUE)

fslr: Uses method by Sled, Zijdenbos, and Evans (1998) (slow)

bc_t1 = fsl_biascorrect(file = base_t1)
FSLDIR='/usr/local/fsl/'; export FSLDIR; sh "${FSLDIR}/etc/fslconf/fsl.sh"; FSLOUTPUTTYPE=NIFTI_GZ; export FSLOUTPUTTYPE; $FSLDIR/bin/fast    -B --nopve --out="/var/folders/1s/wrtqcpxn685_zk570bnx9_rr0000gr/T//Rtmpa4mKOM/file15e6f7f9ff9a7" "/private/var/folders/1s/wrtqcpxn685_zk570bnx9_rr0000gr/T/Rtmpa4mKOM/file15e6f60841581.nii.gz";

ratio = finite_img(n4_t1 / base_t1)
ortho2(n4_t1, ratio, col.y = alpha(hotmetal(), 0.5))

ss_t1 = fslbet(n4_t1, outfile = "SS_Image")
FSLDIR='/usr/local/fsl/'; export FSLDIR; sh "${FSLDIR}/etc/fslconf/fsl.sh"; FSLOUTPUTTYPE=NIFTI_GZ; export FSLOUTPUTTYPE; $FSLDIR/bin/bet2 "/private/var/folders/1s/wrtqcpxn685_zk570bnx9_rr0000gr/T/Rtmpa4mKOM/file15e6f2126f0db.nii.gz" "./SS_Image" 

mask = ss_t1 > 0
ortho2(base_t1, y = mask, col.y = alpha("red", 0.5))

cropped = dropEmptyImageDimensions(ss_t1)
image(cropped, z = floor(dim(cropped)[3]/2), plot.type = "single")

• rgl, misc3d (contour3d function)
• brainR - put on a webpage with some controls

devtools::source_gist("bd40d10afabc503d71e8")

• registration from extrantsr is a general function to do linear/non-linear registration
• uses antsRegistration

ants_reg_flair = registration(
  filename = files["flair"], 
  template.file = n4_t1, 
  typeofTransform = "Rigid")

• Registering the skull-stripped T1-weighted image to the MNI 152 brain image
• Uses Symmetric Normalization (SyN) (B. B. Avants et al. 2008)

template.file = mni_fname(mm = "1", brain = TRUE)
ss_t1_to_mni = registration(
  filename = ss_t1, 
  template.file = template.file, 
  typeofTransform = "SyN", remove.warp = FALSE,
  outprefix = "temp")

reg_flair_to_mni = ants_apply_transforms(
  fixed = template.file, 
  moving = ants_reg_flair$outfile, # registered FLAIR
  interpolator = "Linear",
  transformlist = ss_t1_to_mni$fwdtransforms )

We can develop pipelines/full analyses!

extrantsr::preprocess_mri_within will do inhomogeneity correction, skull strip (or mask), and register to the first scan.

proc_images = preprocess_mri_within(
  files = files[c("t1", "t2", "pd", "flair")], 
  maskfile = ss_t1 > 0)

preprocess_mri_across combines preprocess_mri_within and registration. If you had baseline/follow-up data:

outfiles = gsub("[.]nii", '_process.nii', files)
preprocess_mri_across(
  baseline_files = files[c("base_t1", "base_t2", "base_pd", "base_flair")],
  followup_files = files[c("f_t1", "f_t2", "f_pd", "f_flair")],
  baseline_outfiles = outfiles[c("base_t1", "base_t2", "base_pd", "base_flair")],
  followup_outfiles = outfiles[c("f_t1", "f_t2", "f_pd", "f_flair")],
  maskfile = "Brain_Mask.nii.gz")

• fsl_slicetimer - slice timing correction
• ANTsR::preprocessfMRI
• spm12r
  • slice timing correction
  • realignment - get motion correction parameters
  • co-registration
  • segmentation/normalization to a template

• WhiteStripe (Shinohara et al. 2014)
• Whole brain z-scoring
• Histogram matching
• General standardization methods

• Many methods are being developed for processing neuroimaging in R
• Analysis tools are largely already in R
• These are not standardized nor centralized
• Develop something like BioConductor
  • standard data structures
  • standard data sets
  • NITRC isn't exactly what we want
• GitHub and Neuroconductor